Question How difficult is it to upgrade Intel's fabs to the latest processes, if they get the recipe?

[Failnaught]

Intel has a lot of fab issues, which appears to be giving them indigestion throughout their product design pipeline. It seems like all options are on the table to solve the problem, with a decision possibly coming as soon as early 2021, including using external foundries. But of course nobody has the spare capacity to make all of Intel's new products, and Intel can't afford to idle all of their fabs anyway.

The obvious solution is to develop/buy/license/sell all the engineers' souls for the recipe for a state of the art process. Let's say they managed to do this. How easy would it be to reconfigure their fabs for the new process, and how long would you expect this to take?

By the way, what are the institutional issues that led to Intel manufacturing slipping in such a major way? Prior to 14nm these guys were kicking ass and taking names. 14 nm was late but 14+++++ products still competes favorably with TSMC 7nm in limited ways. It's easy to say "they were on top, and so nobody innovated". That's certainly true on the design front (nobody wants to make a clean sheet chip design if the previous one is still the best), but on the manufacturing side, the need of a better process every 2 years should have been obvious to everyone, even in the C-suite. Does anyone have any insight?

 

[moinmoin]

Intel regularly converts its fab lines between process nodes to keep them all busy. Last time that caused the 14nm shortage since they moved some 14nm fabs to 10nm as planned but 10nm didn't work as expected. It corrected that by moving some back to 14nm and some to 7nm instead, presumably to "accelerate" development on the latter. I don't know if the duration of the downtime of a fab line needed for transitioning it to a different process node is public knowledge.

As for institutional issues that led to Intel manufacturing slipping, common assessment is that Intel tried too much at once for the node jump and failed to make it work as a whole, needing to withdraw some of the proposed changes to try to make it work and repeat the development steps. That the issues keep going on and 7nm appears to be affected as well (after publicly claiming a return to a 2 years cadence no less) points to serious issues in management and execution at Intel in general though.

 

[scannall]

Intel has a lot of fab issues, which appears to be giving them indigestion throughout their product design pipeline. It seems like all options are on the table to solve the problem, with a decision possibly coming as soon as early 2021, including using external foundries. But of course nobody has the spare capacity to make all of Intel's new products, and Intel can't afford to idle all of their fabs anyway.

The obvious solution is to develop/buy/license/sell all the engineers' souls for the recipe for a state of the art process. Let's say they managed to do this. How easy would it be to reconfigure their fabs for the new process, and how long would you expect this to take?

By the way, what are the institutional issues that led to Intel manufacturing slipping in such a major way? Prior to 14nm these guys were kicking ass and taking names. 14 nm was late but 14+++++ products still competes favorably with TSMC 7nm in limited ways. It's easy to say "they were on top, and so nobody innovated". That's certainly true on the design front (nobody wants to make a clean sheet chip design if the previous one is still the best), but on the manufacturing side, the need of a better process every 2 years should have been obvious to everyone, even in the C-suite. Does anyone have any insight?

It's not so much the fabs, though that would be expensive. It's redesigning all your products for new design tools that you're unfamiliar with. Then qualifying them etc. It would set them back years.

 

[AmericanLocomotive]

Retrofitting a fab from one process to another is a huge time and money investment. Global Foundry's Fab 8 in NY took ~3-4 years of construction before it was doing anything. For existing fabs, I'd imagine a changeover from one node to the other would be about a 1-1.5 year endeavor. Also keep in mind that it'd also be a 1-2 year endeavor at the minimum to port an existing functioning architecture to a new node.

 

[Failnaught]

For existing fabs, I'd imagine a changeover from one node to the other would be about a 1-1.5 year endeavor. Also keep in mind that it'd also be a 1-2 year endeavor at the minimum to port an existing functioning architecture to a new node.

It's redesigning all your products for new design tools that you're unfamiliar with. Then qualifying them etc. It would set them back years.

ouch... That's gotta hurt. How tightly integrated are the recipes themselves from the design tools? Does the design software need to know the details of the recipe? I imagine that the software would contain some drawings standard optimized "transistors" and other parts and simple circuits, and the engineer will need to put everything together with software aid? How does it work?

Sorry about the naive questions. I've only done simple nanofab at a university in a materials science way, but I haven't seen any IC design. For us, a complicated device has 16 contacts and with a few material layers stacked together.

 

[Fallen Kell]

Well, it is the death of Intel if they drop their in-house fabs (or even just give up on the high end/cutting edge). Intel is selling off its memory businesses, and if they drop their fabrication, they are now at the mercy of the other global fabrication plants. This also means their designs are being sent to other places to build, which also means they will be stolen much more easily and third parties can/will produce them (we already see counterfeit network chips out of China, the same place that Intel used to fabricate them...).

And while upper management spouts things like Intel "could restore advanced manufacturing to its own factories sometime in the future, if it chooses to ", any engineer who has ever worked with leading edge technology knows for a fact that once you back off trying to stay on the leading edge and give up to outsource, you can never get the talent and and staff training/experience needed to get back into the leading edge, as you are already playing catch-up just to get things working on "current gen", while your competition is working on "current gen + 2" (as they already have the next gen transition process ongoing and their research/development is working 2-3 generations in the future).